This article is a statement of philosophy: how we think about forecasting and why our system is built to react rather than predict. It is educational and general, not personalised investment advice, and nothing here is a recommendation to buy, sell, or hold any security.
Almost everyone selling a market opinion is selling a forecast: where the index goes next, which stock is about to run, what the Fed will do. We are not. Shishin does not predict. It reacts to what the market is actually doing, and this piece explains why we think that is the more honest way to build.
The short version. Trying to predict the stock market is a weak foundation because the future is genuinely uncertain and forecasters are wrong far more often than their confidence suggests. A reactive system defines rules that fire on observable price and regime conditions, and it responds when those conditions appear. We build the second kind. The honest cost is that reacting means you are never early and you take real drawdowns at the turns.
Why prediction is a weak foundation
Many forecasters are careful and sincere. The trouble is that the thing they are forecasting is genuinely uncertain in a way that does not yield to more effort. Markets aggregate the decisions of millions of people reacting to each other, to news that has not happened yet, and to each other’s reactions. That is a reflexive, adaptive system, not a clockwork one, and its next move is not sitting in the present data waiting to be extracted by a smart enough model.
The track record of confident prediction bears this out. As a general observation, year-ahead index targets tend to cluster near a modest positive average and are frequently wrong at the extremes, missing worst in exactly the years that mattered most. The forecasts that would have been worth the most, the ones that called the crash or the melt-up, are the ones the consensus almost never makes, because the consensus is built by people who are penalised more for being alone and wrong than for being in the crowd and wrong. Philip Tetlock spent two decades scoring expert political and economic forecasts and found the average expert did little better than chance, and that the most confident and most famous often did worse than a simple extrapolation algorithm. The future is not unknowable in every respect. But building a system on a foundation of “we know what happens next” is building on sand.
Predicting versus reacting
The distinction we care about is not subtle, but it is easy to blur, so it is worth stating cleanly.
A prediction is a claim about the future: this stock will go up, the market will fall next quarter, this is the bottom. It commits you before the evidence arrives. If it is wrong, you were positioned for a world that did not happen, and you find out only after the fact.
A reaction is a rule triggered by something already true: when price does this observable thing, take that action. It commits you only after the condition appears in the data you can see. It does not claim to know what tomorrow holds. It claims only to have decided, in advance and in writing, what it will do when specific, observable circumstances arrive. The judgement is spent on the rule, not on the guess.
Both approaches involve rules and both involve risk. The difference is where the uncertainty lives. The forecaster stakes the outcome on being right about the future. The reactor stakes it on the rule being sound over many occurrences, and lets the future reveal itself before acting. One of these is a bet on foresight. The other is a bet on discipline.
Why reacting is more robust
A reactive rule degrades gracefully in a way a forecast does not. When a forecast is wrong, the whole position built on it is wrong at once. A reactive rule that stops working simply stops firing, or fires and gets cut by its own exit, and the damage is bounded by the risk you set before you entered rather than by the size of your conviction. To survive, the system needs only to respond consistently and keep its losses small when a response does not pan out.
Reacting also sidesteps the forecaster’s worst failure mode: commitment to a view. Once you have publicly predicted a direction, the evidence against it becomes something to argue with rather than something to act on. A rule that keys off observable conditions has no ego in the trade. It does not need the market to vindicate a call. It just reads what is in front of it and applies the same logic it applied yesterday. That is a big part of why we prefer it. It is harder to fool yourself with.
How Shishin reacts
This is the whole design of the system, not a marketing gloss. Shishin has two reactive layers, and neither one forecasts.
The first is a regime classifier that reads current market conditions, breadth, trend, volatility, the state of the tape as it is now, and routes to the engine suited to that environment. In a healthy, trending market it leans on momentum through Suzaku. As conditions deteriorate it rotates toward the quality of Genbu, the defensive posture of Byakko, or simply toward holding back. When a beaten-down tape starts to mend it leans on the large-cap recovery of Seiryu. Crucially, the classifier makes no call on the top or the bottom and never says “a correction is coming.” It reads what the market already is and responds to that. It reacts to the regime it can observe, and it will be late to every turn, by construction, because it waits for the condition to be real before it moves. That lateness is the price of not guessing, and we pay it on purpose. We wrote more about keeping the live machine faithful to that design in running what was tested.
The second layer is the ranked board, which reacts to price. Each morning the model scores its universe on observable characteristics and publishes a ranking. A name rises on the board because of what its price and its setup are doing now, not because anyone forecast that it will go up. The board is a reading of present strength, not a prophecy. It is also, deliberately, research rather than advice, a reaction to the tape made visible. It never tells anyone what to buy, and what a reader does with it is entirely their own decision.
Put together, the honest one-line description of the system is this: it knows what it will do when the market does something observable, and it does not pretend to know anything else. That is the entire claim.
The honest limit: reacting means you are never early
A reactive system buys a real, permanent disadvantage, and we would be dishonest to hide it. If you wait for a condition to be observable before you act, you are by definition never early. You will not catch the exact bottom, because at the bottom the observable conditions still look terrible. You will not sidestep the start of a decline, because at the top the tape still looks strong. The regime classifier turns defensive after the deterioration is visible, not before it, which means you ride some of the way down before the system reacts. That is the direct, unavoidable cost of refusing to guess, and we are not working to fix it.
Concretely, this shows up as drawdowns at the turns. A momentum posture held into a market that rolls over will give back gains before the regime read flips, and momentum as a strategy has a well-documented failure mode at exactly those inflection points. A forecaster who happened to call the turn correctly would look far better than us in that window. Sometimes one will. We accept that trade because the forecaster who calls the turn correctly is drawn from the same population that calls it incorrectly most of the time, and we would rather be reliably a little late than occasionally brilliant and frequently ruined. A rule that reacts can be wrong too. It is just wrong in bounded, survivable ways instead of catastrophic ones.
Judge the posture, do not take our word for it
The claim that a system reacts rather than predicts is easy to assert and worth very little unless you can check it. So we make it checkable. The daily board and the regime read are published, and the live, paper-traded results, winners and losers alike, sit on the public track record with the misses shown next to the wins. The signals and NAV are committed as a tamper-evident, timestamped chain you can audit on the verification page, so a reaction cannot be quietly reframed as a prediction after the outcome is known. Whether a given signal has any edge is a separate and fair question, one we take up honestly in do stock signals actually work. The point of this piece is narrower: we are not asking you to trust a forecast, because we do not make one.
The usual honest caveats still apply and always will. Our backtest is a five-year hypothesis, not a promise, and the live record is still young. Reacting instead of forecasting keeps a system honest about what it is doing, which is responding to observable conditions with rules set in advance, and never claiming to see around the next corner.
Sources & further reading
- Tetlock, P. E. (2005). Expert Political Judgment: How Good Is It? How Can We Know? Princeton University Press.
- Taleb, N. N. (2007). The Black Swan: The Impact of the Highly Improbable. Random House.
- Running what was tested on keeping the live system faithful to the rules it was measured on.
- Momentum investing explained, including the failure mode a reactive system rides through at the turns.
- Do stock signals actually work for how to judge whether any signal, ours included, has real edge.
- The live, paper-traded results on the track record, and the tamper-evident log on the verification page.